Bug when evaluating model formula in FGR()

[ellessenne]

Hi,

I think I have found a bug when using the FGR() function and when the model formula contains other functions (e.g., splines ns() or polynomials poly()):

library(riskRegression)
#> riskRegression version 2023.12.21
library(prodlim)
library(survival)
library(cmprsk)
d <- prodlim::SimCompRisk(100)
f1 <- FGR(Hist(time, cause) ~ X1 + X2, data = d)
#> Argument cause missing. Analyse cause: 1
print(f1)
#> 
#> Right-censored response of a competing.risks model
#> 
#> No.Observations: 100 
#> 
#> Pattern:
#>          
#> Cause     event right.censored
#>   1          47              0
#>   2          25              0
#>   unknown     0             28
#> 
#> 
#> Fine-Gray model: analysis of cause 1 
#> 
#> Competing Risks Regression
#> 
#> Call:
#> FGR(formula = Hist(time, cause) ~ X1 + X2, data = d, cause = "1")
#> 
#>     coef exp(coef) se(coef)    z p-value
#> X1 1.232      3.43    0.294 4.19 2.7e-05
#> X2 0.913      2.49    0.154 5.93 3.0e-09
#> 
#>    exp(coef) exp(-coef) 2.5% 97.5%
#> X1      3.43      0.292 1.93  6.10
#> X2      2.49      0.401 1.84  3.37
#> 
#> Num. cases = 100
#> Pseudo Log-likelihood = -170 
#> Pseudo likelihood ratio test = 40.6  on 2 df,
#> 
#> Convergence: TRUE

library(splines)
FGR(Hist(time, cause) ~ X1 + ns(X2, df = 3), data = d)
#> Error in str2lang(termtext): <text>:2:0: unexpected end of input
#> 1: X1+ns(X2
#>    ^

FGR(Hist(time, cause) ~ X1 + poly(X2, degree = 2), data = d)
#> Error in str2lang(termtext): <text>:2:0: unexpected end of input
#> 1: X1+poly(X2
#>    ^

Note that no error happens if I use I() instead, I think it may due to the , character breaking the str2lang call?

FGR(Hist(time, cause) ~ X1 + X2 + I(X2^2), data = d)
#> Argument cause missing. Analyse cause: 1
#> 
#> Right-censored response of a competing.risks model
#> 
#> No.Observations: 100 
#> 
#> Pattern:
#>          
#> Cause     event right.censored
#>   1          47              0
#>   2          25              0
#>   unknown     0             28
#> 
#> 
#> Fine-Gray model: analysis of cause 1 
#> 
#> Competing Risks Regression
#> 
#> Call:
#> FGR(formula = Hist(time, cause) ~ X1 + X2 + I(X2^2), data = d, 
#>     cause = "1")
#> 
#>           coef exp(coef) se(coef)     z p-value
#> X1      1.2217      3.39   0.2964 4.121 3.8e-05
#> X2      0.8504      2.34   0.1724 4.932 8.1e-07
#> I(X2^2) 0.0638      1.07   0.0923 0.691 4.9e-01
#> 
#>         exp(coef) exp(-coef)  2.5% 97.5%
#> X1           3.39      0.295 1.898  6.07
#> X2           2.34      0.427 1.669  3.28
#> I(X2^2)      1.07      0.938 0.889  1.28
#> 
#> Num. cases = 100
#> Pseudo Log-likelihood = -170 
#> Pseudo likelihood ratio test = 41  on 3 df,
#> 
#> Convergence: TRUE

I know I can pre-compute the spline expansion, so that would be a workaround, but I thought about reporting this issue anyway. Hope this helps!

Alessandro

[tagteam]

thank you! yes, this is not supported. will put a stop criterion in unless you want to fix the bug?

best T

Alessandro Gasparini @.***> writes:

Hi,

I think I have found a bug when using the FGR() function and when the model formula contains other functions (e.g., splines ns() or polynomials poly()):

library(riskRegression)
#> riskRegression version 2023.12.21
library(prodlim)
library(survival)
library(cmprsk)
d <- prodlim::SimCompRisk(100)
f1 <- FGR(Hist(time, cause) ~ X1 + X2, data = d)
#> Argument cause missing. Analyse cause: 1
print(f1)
#> 
#> Right-censored response of a competing.risks model
#> 
#> No.Observations: 100 
#> 
#> Pattern:
#>          
#> Cause     event right.censored
#>   1          47              0
#>   2          25              0
#>   unknown     0             28
#> 
#> 
#> Fine-Gray model: analysis of cause 1 
#> 
#> Competing Risks Regression
#> 
#> Call:
#> FGR(formula = Hist(time, cause) ~ X1 + X2, data = d, cause = "1")
#> 
#>     coef exp(coef) se(coef)    z p-value
#> X1 1.232      3.43    0.294 4.19 2.7e-05
#> X2 0.913      2.49    0.154 5.93 3.0e-09
#> 
#>    exp(coef) exp(-coef) 2.5% 97.5%
#> X1      3.43      0.292 1.93  6.10
#> X2      2.49      0.401 1.84  3.37
#> 
#> Num. cases = 100
#> Pseudo Log-likelihood = -170 
#> Pseudo likelihood ratio test = 40.6  on 2 df,
#> 
#> Convergence: TRUE

library(splines)
FGR(Hist(time, cause) ~ X1 + ns(X2, df = 3), data = d)
#> Error in str2lang(termtext): <text>:2:0: unexpected end of input
#> 1: X1+ns(X2
#>    ^

FGR(Hist(time, cause) ~ X1 + poly(X2, degree = 2), data = d)
#> Error in str2lang(termtext): <text>:2:0: unexpected end of input
#> 1: X1+poly(X2
#>    ^

Note that no error happens if I use I() instead, I think it may due to the , character breaking the str2lang call?

FGR(Hist(time, cause) ~ X1 + X2 + I(X2^2), data = d)
#> Argument cause missing. Analyse cause: 1
#> 
#> Right-censored response of a competing.risks model
#> 
#> No.Observations: 100 
#> 
#> Pattern:
#>          
#> Cause     event right.censored
#>   1          47              0
#>   2          25              0
#>   unknown     0             28
#> 
#> 
#> Fine-Gray model: analysis of cause 1 
#> 
#> Competing Risks Regression
#> 
#> Call:
#> FGR(formula = Hist(time, cause) ~ X1 + X2 + I(X2^2), data = d, 
#>     cause = "1")
#> 
#>           coef exp(coef) se(coef)     z p-value
#> X1      1.2217      3.39   0.2964 4.121 3.8e-05
#> X2      0.8504      2.34   0.1724 4.932 8.1e-07
#> I(X2^2) 0.0638      1.07   0.0923 0.691 4.9e-01
#> 
#>         exp(coef) exp(-coef)  2.5% 97.5%
#> X1           3.39      0.295 1.898  6.07
#> X2           2.34      0.427 1.669  3.28
#> I(X2^2)      1.07      0.938 0.889  1.28
#> 
#> Num. cases = 100
#> Pseudo Log-likelihood = -170 
#> Pseudo likelihood ratio test = 41  on 3 df,
#> 
#> Convergence: TRUE

I know I can pre-compute the spline expansion, so that would be a workaround, but I thought about reporting this issue anyway. Hope this helps!

Alessandro

-- 7LL-5 Don’t compare your life with others, you have no idea what their journey is all about.

[ellessenne]

I can have a look into it!